Comparative mathematical and data-driven models for simulating the performance of forward osmosis membrane under different draw solutions

Viet, N, D., Jang, A.

Mathematical models have frequently been used to simulate system performance, including forward osmosis (FO)-related processes. More recently, Industry 4.0 has increased in importance, and artificial intelligence (AI) technology has proven to be effective at assisting decision-making across a range of industries, including membrane processes. This study, for the first time, proposes a comparison of the accuracy of predictions for water permeability under various types of draw solutions using conventional models and AI-based models. The results showed that the AI-based models, including artificial neural networks (NN) (2 layers; 10 neurons) and adaptive neuro-fuzzy inference system (NFIS) (with grid partition; membership functions = 5) performed significantly better than the conventional mathematical intermediate blocking (CIB) model, with the coefficient values (R) of 0.974, 0.994, and 0.963, respectively. The corresponding errors of these simulations were 0.04, 0.70, and 0.97 LMH, indicating that the AI technique was better suited than the mathematical CIB model for simulating FO-related system performance. Additionally, using AI to forecast system performance is timely (i.e., considerably quicker than developing a sophisticated mathematical model) and cost-effective (i.e., reducing the expense for analysis of environmental factors for CIB model construction). The authors also successfully constructed an application for end-users for better and simpler experiences with modeling and simulation, even for those with varying knowledge in the field.